Software Technology Research Laboratory, Mobile Agents in Distributed Environments: Principles and Paradigms Kevin Jones 20 th January 2005

Software Technology Research Laboratory, This is part of a series of seminars offered to introduce the notion of security in Agent Distributed Systems. This is aimed to be a general overview of the principles behind, and application of Mobile Agents. The scope of this presentation includes: What are Agents (reminder)? What are Mobile Agents, how do they differ and present more complex security issues? Example Agent implementation utilising the Jade Platform. Agent Security Techniques. 2

Software Technology Research Laboratory, What is an Agent? (reminder) 3

Software Technology Research Laboratory, " The term agent is used to represent two orthogonal concepts. The first is the agent's ability for autonomous execution. The second is the agent's ability to perform domain oriented reasoning.“ The MuBot Agent "An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through effectors." The AIMA Agent [Russell and Norvig 1995, page 33] "Autonomous agents are computational systems that inhabit some complex dynamic environment, sense and act autonomously in this environment, and by doing so realize a set of goals or tasks for which they are designed." The Maes Agent [Maes 1995, page 108] "Let us define an agent as a persistent software entity dedicated to a specific purpose. 'Persistent' distinguishes agents from subroutines; agents have their own ideas about how to accomplish tasks, their own agendas. 'Special purpose' distinguishes them from entire multifunction applications; agents are typically much smaller." The KidSim Agent [Smith, Cypher and Spohrer 1994] Intelligent agents continuously perform three functions: perception of dynamic conditions in the environment; action to affect conditions in the environment; and reasoning to interpret perceptions, solve problems, draw inferences, and determine actions. The Hayes- Roth Agent [Hayes-Roth 1995] "Intelligent agents are software entities that carry out some set of operations on behalf of a user or another program with some degree of independence or autonomy, and in so doing, employ some knowledge or representation of the user's goals or desires." The IBM Agent "... a hardware or (more usually) software-based computer system that enjoys the following properties: autonomy: agents operate without the direct intervention of humans or others, and have some kind of control over their actions and internal state; social ability: agents interact with other agents (and possibly humans) via some kind of agent-communication language; reactivity: agents perceive their environment, (which may be the physical world, a user via a graphical user interface, a collection of other agents, the INTERNET, or perhaps all of these combined), and respond in a timely fashion to changes that occur in it; pro-activeness: agents do not simply act in response to their environment, they are able to exhibit goal-directed behavour by taking the initiative." The Wooldridge­p;Jennings Agent [Wooldridge and Jennings 1995, page 2] "Software agents are programs that engage in dialogs [and] negotiate and coordinate transfer of information.“The SodaBot Agent [Michael Coen ] "Autonomous agents are systems capable of autonomous, purposeful action in the real world.“The Brustoloni Agent [Brustoloni 1991, Franklin 1995, p. 265] “an autonomous process capable of reacting to, and initiating changes in, its environment, possibly in collaboration with users and other agents” [Jennings and Woolridge, 1998] 4

Software Technology Research Laboratory, It would seem that there is no single definition of an agent. The term is widely used in many spheres including: Distributed Computing Artificial Intelligence (AI) Network Management Human Computer Interaction Manufacturing 5

Software Technology Research Laboratory, ….so lets start at the beginning!! An Agent is a “Thingy” 6

Software Technology Research Laboratory, Most literature regarding Agents introduce a definition of what an agent is in terms of its properties. So for our purpose, an Agent possesses following properties: AutonomousExercises control over its own actions Goal OrientedDoes more than just react to the environment, has an agenda. ReactiveCan sense and respond to changes in its environment FlexibleOrder of actions is not fixed. CommunicativeCommunicates with other agents (or people) Adaptive / Learning Can change behaviour based on experience MobileCan ‘migrate’ from one host to another 7

Software Technology Research Laboratory, For our purposes, one of the more appropriate definitions to our use of Agents reads: “An Autonomous Agent is a system situated within, and part of an environment, that senses that environment and acts on it, over time, in pursuit of its own agenda and so as to effect what it senses in the future.” Franklin & Graesser {Is it an Agent, or just a program? – A Taxonomy for Autonomous Agents}

Software Technology Research Laboratory, There are many ‘Agent Frameworks’ available to produce Agent Systems, whilst many of these are based on the ‘Java’ Programming language (for portability reasons) this is not always necessarily the case. Ara TclC, Tcl ffMain TclPerl, Java, Tcl Ajanta, Jade, MoleJavaTelescript APRILAPRIL Scripting Language 9

Software Technology Research Laboratory, In 1996 The Foundation for Intelligent Physical Agents (FIPA) was formed to produce software standards for heterogeneous and interacting agents and agent-based systems. This gave rise to more agent platforms, comparable to this standard: APRIL Agent Platform [non Java based, uses the ‘Agent PRocess Interaction Language’] Comtec Agent Platform [requires SL2 as content language and provides minimum documentation] FIPA-OS [only theoretical mobility – the mobility is still a prototype] Grasshopper JACK intelligent Agents [does not support mobility] Java Agent Services (JAS) Jade / LEAP Zeus [no mobility] 10

Software Technology Research Laboratory, Why Mobile Agents? 11

Software Technology Research Laboratory, Mobile Agents have many advantages: Reducing Network Load - Reduce network interactions E –Commerce – Purchasing Agents can easily be dispatched to find, barter and buy goods. Network Management & Asynchronous Execution - Mobile devices (such as PDA’s) can be turned off after an agent has migrated and the agent is still able to perform tasks within a network. It is then possible (if an ‘agent-garage’ is in place) for the agent to return to the device upon reconnection to the network. Resource Utilisation - Processing capabilities can be shared or transferred to devices more equipped to the task. Autonomy / Intelligence – Agents can make decisions, communicate with their environment and process information) 12

Software Technology Research Laboratory, Example Scenarios of Mobile Agents could include: Searching Multiple Request from Large Database Infrastructures Avoidance of Network Congestion and Performance Degradation Shopping Agents and Booking Agents Agents Calendar Agents Tracking Agents (Mobile Shadow – University of Salzburg) [ ] 13

Software Technology Research Laboratory, 14 Peer to Peer ModelMobile Agent Model Peer-to-peer style communication

Software Technology Research Laboratory, 15

Software Technology Research Laboratory, 16

Software Technology Research Laboratory, 17 An Agent is sent to find the best price for flowers. First it discovers an E-bazaar may be cheaper so migrates to this container. Once at the Bazaar, the agent interacts with the environment (i.e. other agents) to bid for the flowers. Once the price exceeds a reasonable price calculated by the agent it withdraws from the bidding. It can then “Autonomously” migrate to a flower-shop container to purchase the flowers from there before returning home to report the purchase. [ Here the agent has knowledge of its location, its’ environment and has been able to calculate an appropriate price, negotiated with other agents, processed this information and found an alternative source for goods. ] Agent Migrates to E-Bazaar to purchase flowers

Software Technology Research Laboratory, Mobile Agents (“Migration”) 18

Software Technology Research Laboratory, Mobile Code is not a new concept! Even as early as the 1960’s Remote Job Entry Terminals were being used to submit programs to a central computer. The modern implementation is Java applets, these can be considered as ‘mobile code’, downloaded via a web browser to execute on a local machine. Mobile Agent technology, however, takes this even further, allowing complete mobility of cooperating applications among supporting platforms to form a large- scale, loosely coupled distributed system. 19

Software Technology Research Laboratory, Timeline 20

Software Technology Research Laboratory, It is generally seen that there are TWO distinct types of migration Weak Migration – is the ability of an agent to migrate its code to another host, this code may be accompanied by some initialisation data, but no migration of execution state is involved. Strong Migration – Strong mobility allows an agent to transfer both the code and the execution state to another environment. It is also possible to continue execution from the point of migration. NOTE: To implement Strong Migration requires a modified version of the Java Virtual Machine! [Jessica2] 21

Software Technology Research Laboratory, JADE JADE (Java Agent DEvelopment Framework) is a software framework to develop agent-based applications. Originally developed in 1999 it has been updated many times since and is now FIPA compliant. JADE can be considered an agent middleware that implements an Agent Platform and a development framework. The agent platform can be distributed on several hosts. Only one Java application, and therefore only one Java Virtual Machine (JVM), is executed on each host. Each JVM can be considered a container of agents. 22

Software Technology Research Laboratory, 23

Software Technology Research Laboratory, 24

Software Technology Research Laboratory, Utilising the Jade Agent platform as an example: Agents operate within a ‘Container’ hosted on an Agent Platform. Platforms may be distributed, containing multiple containers Messages are passed, allowing Agents to communicate with other agents within a container. Although it is possible to pass messages across containers (i.e. normal network load), an Agent can also migrate to a new container within the same platform and utilise local message passing, thus minimising network bandwidth usage. 25

Software Technology Research Laboratory, 26 Agent platform example for the Jade Environment

Software Technology Research Laboratory, Example Agent (Jade). 27

Software Technology Research Laboratory, Jade Agents are created by extending the jade.core.Agent class and redefining the setup() method Agents are identified by a unique name at a container address, this is know as an AID. Jade Agents respond to behaviours. Each behaviour is created by extending the jade.core.behaviours.Behaviour class 28

Software Technology Research Laboratory, Behaviour Types “One – Shot” – Complete immediately and are executed only once. “Cyclic” – Never Completes, executed same operation at each invocation “Complex” – Actions may vary depending on the state of the agent 29

Software Technology Research Laboratory, Jade Mobility “doMove()” – the doMove() command can be called to force an agent to migrate to a given environment. “beforeMove()” – the beforeMove() will be used to ensure that an agent is prepared to migrate. afterMove()” – within the Jade framework, the execution continues from the afterMove() method once an agent has completed its migration. 30

Software Technology Research Laboratory, Flower Shop Agent This is an example of a flower shop agent. Although, for demonstration purposes only, it gives a good understanding of the ability of an agent. In this case to migrate, perform a given task then migrate back. 31

Software Technology Research Laboratory, Agent Security. 32

Software Technology Research Laboratory, Lack of Agent Security is one of the reasons why Agent technology is not being brought to the mainstream. Both Agent Technology and thus, their associated security issues are classed as an emerging technology and in somewhat of an infancy stage. Much research is being carried out both by academic institutions and within commercial environments to solve many of the security issues. The Software Technology Research Laboratory (De Montfort University) is no exception. Currently the (Mobile) Agent Security Research group consists of: Professor Hussein ZedanFrancois Siewe Dr Antonio CauHelge Janicke Kevin Jones 33

Software Technology Research Laboratory, 4 Key areas of Mobile Agent Security: Protecting Hosts from Agents Protecting Hosts from Hosts Protecting Agents from Hosts Protecting Agents from Agents 34

Software Technology Research Laboratory, Much research has already been carried out by many institutions to review the issue of ‘Host Protection’. Attacks on Hosts can include: Modification, deletion, hosts system files. Denial of Service (DoS) Invasion of Privacy, theft or undesirable access to data on the host system. Masquerading, a malicious agent claims the identity of another (possibly authorised/ trusted) agent. Snooping, agents can monitor the behaviour of a host and gain information about its dealings 35

Software Technology Research Laboratory, Many solutions offered to solve these issues are based on serving as a deterrent, although, the combination of many or all of these techniques offers some hope for future security. Solutions include: Digitally Signed Agents / Authentication, thus an audit log would provide a mechanism for compensation if a malicious or fraudulent act were undertaken. Potentially allows for agent code to be crosschecked with an MD5 hash to detect manipulations in transit. Sandbox / Blackbox Theory, standard Java toolkit to allow for untrusted code to be executed in an untrusted domain with privileges defined in a ‘policy’. State Appraisal, state is monitored and predicted to monitor for potentially harmful actions. 36

Software Technology Research Laboratory, Protecting the Agent itself has proved to be much trickier!! 37

Software Technology Research Laboratory, Agent V Host A host system has control over the execution state of an agent It controls all of the Input / Output of both the agent and itself. All interpreted languages (such as Java based platforms – Jade) must be executed by a Virtual Machine, this VM may be modified to become malicious towards an agent. 38

Software Technology Research Laboratory, Agent’s are vulnerable to attacks such as: Access Denial – a host may prevent the agent from executing or block the communication channels that allow message passing or migration. Unauthorised access/modification of data – the data contained within an agent (credit card information for example) may only wish to be shared with certain hosts on a multi-hop migration. This data may be vulnerable to theft or modification. [Example – booking agent, best offer] Modification of code – If an host modifies the code of an agent this could then deem the agent to be malicious. Masquerade – a host may masquerade as a different host to trick an agent into believing it is on a ‘trusted’ host and thus extract sensitive information. 39

Software Technology Research Laboratory, Agent V Agent Agents may execute in the same container (Virtual Machine) and many communicate and interact with each other. Thus, malicious Agents may extract sensitive data from or cause harm to other agent executing within the same container. 40

Software Technology Research Laboratory, Current Solutions: Itinerary Recording, allowing an agents path to be recorded for later inspection and repudiation. Similarly ‘execution tracing’ can be used. Encrypted Functions, the functions of an agent are encrypted so that the host can execute said function without deciphering the original code. Obfuscated Code, the code is scrambled, such that the original meaning is difficult to follow – this relies heavily on time for protection (i.e. the code is more difficult to follow and thus takes longer to manipulate) protecting the agent for a limited time. State Appraisal, Agents monitor their current state and monitor at which state they expect to be in next. Any abnormalities can be monitored. 41

Software Technology Research Laboratory, Current Solutions (continued): Authentication / Access Control, Agents should authenticate themselves to perform a specific operation. Policies can be generated to govern the access. Formal policies can be utilised to specify security constraints from a given set of system requirements. SaNTA SaNTa is designed to ease the development of security aware agent systems using a formal approach. 42

Software Technology Research Laboratory, 43

Software Technology Research Laboratory, Future Work Develop ‘Vigilant Agents’ and thus the requirements for agent security Utilise the SaNTA Development Language to produce more advanced agent systems with secure vigilant agents. 44